Unverified50% confidenceBenchmarkExact time
在ELE测试中等推理强度下，Sol的成本约为Faber-5的1/4，Terra和Luna成本仅约1/16且得分仍超过Faber-5
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/24/2026
First Seen
Valid until: 10/22/2026
Sources
GPT-5.6 Three-Model Showdown: Sol vs Terra vs Luna — Which One Should You Pick?
bilibili麦冬AI实验室7/11/2026
Related Claims
Unverified在 Frontier Code 准确率与成本基准测试中，低档位花费约 5 美元可达约 11% 得分，而 Opus 4.8 Max 花费约 11 美元才拿到相同得分69% similarUnverifiedIn the ExploitBench test, Sol scored approximately 76%, slightly behind Fable 5's 78%, but Sol consumed only about 120,000–130,000 tokens while Fable 5 used approximately 350,000 tokens.65% similarUnverifiedThe three names Sol, Terra, and Luna represent capability tiers rather than generation numbers, with Sol being the flagship and most powerful, Terra targeting high-frequency daily work, and Luna focusing on cost-effectiveness.62% similarUnverifiedBenchmark results show that Sol and Sol Ultra outperform competitor Misos 5 in most areas, and Tara outperforms Fable 5.62% similarUnverifiedOpus 5's pass rate at Low Effort is approximately 60%, surpassing Opus 4.8's 59% at the Max level, for which the per-task cost is as high as $13.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/611087API
curl https://kongchang.com/api/v1/knowledge/claims/611087MCP
get_claim(id=611087)